A novel framework using particle swarm optimization and long short-term memory networks for stock market forcasting

Area of Science:

• Quantitative Finance
• Computational Intelligence
• Natural Language Processing

Background:

• Stock price forecasting is complex due to market nonlinearity and sentiment.
• Traditional models struggle with the dynamic nature of financial markets.
• Accurate forecasting is crucial for investment strategies.

Purpose of the Study:

• To develop an intelligent hybrid framework for improved stock price forecasting.
• To integrate Long-Short-Term Memory (LSTM) networks with metaheuristic optimization.
• To incorporate financial sentiment analysis for enhanced predictive accuracy.

Main Methods:

• Evaluated multiple LSTM network architectures (single, two, three layers).
• Employed metaheuristic algorithms (Particle Swarm Optimization, Gray Wolf Optimization, Artificial Rabbit Optimization) for hyperparameter tuning.
• Integrated financial sentiment features using the FinBERT model.

Main Results:

• Particle Swarm Optimization (PSO) outperformed other optimization techniques.
• The hybrid SEN-PSO-LSTM model demonstrated superior forecasting performance across diverse stocks.
• Sentiment analysis integration significantly boosted accuracy in volatile market conditions.

Conclusions:

• Combining deep learning (LSTM), hyperparameter optimization (PSO), and sentiment analysis (FinBERT) is effective for stock forecasting.
• The proposed framework offers a robust solution for navigating complex financial markets.
• This approach enhances predictive accuracy, particularly under market volatility.